220 likes | 410 Views
Operating Systems CMPSCI 377 Lecture 11: Memory Management. Emery Berger University of Massachusetts, Amherst. Course Outline. Processes & Threads CPU Scheduling Synchronization & Deadlock Memory Management File Systems & I/O Distributed Systems. Today: Memory Management. Terminology
E N D
Operating SystemsCMPSCI 377Lecture 11: Memory Management Emery Berger University of Massachusetts, Amherst
Course Outline • Processes & Threads • CPU Scheduling • Synchronization & Deadlock • Memory Management • File Systems & I/O • Distributed Systems
Today: Memory Management • Terminology • Addresses • Uniprogramming • Multiprogramming • Relocation • Allocation • Policies • Fragmentation • Compaction • Swapping
Memory Management • Where in memory is executing process? • How do we allow multiple processes to share main memory? • What’s an address and how is one interpreted?
Background:Computer Architecture • Program executable on disk • OS loads program into memory • CPU fetches instructions & data from memory while executing program
Memory Management:Terminology • Segment: chunk of memory assigned to process • Physical address: real address in memory • Virtual address: address relative to start of process’s address space
Where Do Addresses Come From? Instruction & data addresses • Compile-time: • Exact physical location in memory starting from fixed position k • Load-time: • OS determines process’s starting position, fixes up addresses • Execution time: • OS can place address anywhere in physical memory
Uniprogramming Only one program at a time) simplifies memory management • OS gets fixed region of memory (e.g., highest) • One process at a time • Load at address 0 • Executes in contiguous memory • Compiler generates physical addresses • Max address = memory size – OS size • OS protected from process by checking addresses
Example: Uniprogramming • Simple – but no overlap of I/O, computation
Multiprogramming Requirements • Transparency • No process aware memory is shared • Process has no constraints on physical memory • Safety • Processes cannot corrupt each other • Or the OS • Efficiency • Performance not degraded due to sharing
Relocation • Put OS in high memory • Process starts at 0 • Max addr = memory size – OS size • Load process by allocating contiguous segment for process • Smallest addr = base, largest = limit
Static Relocation • At load time, OS adjusts addresses in process to reflect position in memory • OS cannot move process after relocation
Dynamic Relocation • Hardware adds relocation register (base) to virtual address to get physical address • Hardware compares address with limit register (addr must be less than base) • Test fails ) trap
Relocation, Pros & Cons • Advantages • OS can easily move process during execution • OS can allow process to grow over time • Simple, fast hardware • Two special registers, add, & compare • Disadvantages • Slows everything (add on every reference) • Can’t share memory between processes • “Text” section • Process limited to physical memory size • Degree of multiprogramming limited • All memory of active processes must fit in memory • Complicates memory management
Relocation, Properties • Transparency • Processes largely unaware of sharing • Safety • Each memory reference checked • Efficiency • Memory checks fast if done in hardware • But: if process grows, may have to be moved (SLOW)
Memory Allocation • As processes enter system, grow & terminate, OS must track available and in-use memory • Can leave holes • OS must decide where to put new processes
Memory Allocation Policies • First-fit: • Use first hole in which process fits • Best-fit: • Use smallest hole that’s large enough • Worst-fit: • Use largest hole • What’s best?
Fragmentation • Fragmentation = % memory unavailable for allocation, but not in use • External fragmentation: • Caused by repeated unloading & loading • Many small holes • No contiguous chunk large enough • Internal fragmentation: • Space inside process allocations • Unavailable to other processes
Compaction • Can make space available by compacting process space • Eliminate holes
Compaction Example • Issues • Amount of memory moved • Size of created block • Other choices?
Alternative: Swapping • Swapping = copy process to disk, release all memory • When process active, must reload • Static relocation: same position(!) • Dynamic relocation: ok • Compaction simplified if swapping supported • But not supported in modern OS’s • Interaction with scheduling?
Summary • Processes must reside in memory to execute • Generally use virtual addresses • Translated to physical addresses before accessing memory • Segmentation: • Allows processes to share memory • Expensive to grow over time • Swapping: • Total being used by all can exceed main memory • Increases context switch time